Counterintuitive outcome of 'p' with numeral prefix

[lichray]

From Walter Alejandro Iglesias wai@roquesor.com

When you yank a word and paste it with 'p' using a numeric prefix, the first character of the string yanked in the last occurrence is pasted at the beginning of the entire string.

In the following example I copied the word "Hello" with the command 'ye' and pasted it on the same line (leaving a space in between to make the example more readable) first using the command '2p' then using '3p':

Hello HHelloHello HHelloHelloHello

In the code, numeral prefixes are handled with a for loop (file vi/v_put.c). The diff at bottom partially solves the issue, it just removes the following line in the v_put() function:

vp->p_start = vp->m_final;

But, I guess that line must be there for a reason ;-). I said "partially" because even after you remove that line the issue is still reproducible when you paste the word in an empty line. For example:

Hello
HHelloHello

--- vi/v_put.c  Thu Apr 24 15:13:56 2025
+++ vi/v_put.c.orig Thu Apr 24 15:13:41 2025
@@ -77,6 +77,7 @@
    {
        if (F_ISSET(vp, VC_BUFFER) ? &vp->buffer : NULL,
            vp->p_start, &vp->m_final, 1))
            return (1);
+       vp->p_start = vp->m_final;
        if (INTERP(vsp))
            return (1);
    }

[lichray]

Historical notes

nvi2/common/put.c

Lines 137 to 150 in 16de509

	* !!!
	* In the historical 4BSD version of vi, character mode puts within
	* a single line have two cursor behaviors: if the put is from the
	* unnamed buffer, the cursor moves to the character inserted which
	* appears last in the file. If the put is from a named buffer,
	* the cursor moves to the character inserted which appears first
	* in the file. In System III/V, it was changed at some point and
	* the cursor always moves to the first character. In both versions
	* of vi, character mode puts that cross line boundaries leave the
	* cursor on the first character. Nvi implements the System III/V
	* behavior, and expect POSIX.2 to do so as well.
	*/
	rp->lno = lno;
	rp->cno = len == 0 ? 0 : sp->cno + (append && tp->len ? 1 : 0);

Observation

'2P' seems to work "fine."

Discussion

The following discussion of how it's implemented in OpenBSD's ksh vi-mode is interesting:

https://marc.info/?t=174527132800001&r=1&w=2

Take a look especially to this message:

https://marc.info/?l=openbsd-tech&m=174551571311274&w=2

It's very likely that whoever designed or corrected the vi mode in
OpenBSD's ksh realized what NilsOla Nilsson pointed out, and improved
those commands. However Ingo Schwarze end opting for the behavior you
see in vim, bash or ksh93 from ports.

[lichray]

From Walter Alejandro Iglesias

suppose we want to paste "Hello" two times times with '2p'.
Currently, this takes to the code the following steps (roughly):

1. Load "Hello" to the buffer.
2. Decide where to move the cursor.
3. Paste "Hello" and move the cursor.
4. Load second occurrence of "Hello" to the buffer.
5. Decide where to move the cursor.
6. Paste the second occurrence of "Hello" and move the cursor.

[...] I moved the loop with the counter to the put() function
itself, which means that all occurrences of "Hello" are loaded into the
buffer before the cursor movement is decided. The '2p' command above is
reduced to:

1. Load "HelloHello" to the buffer.
2. Decide where to move the cursor.
3. Paste "HelloHello" and move the cursor.